A central goal of Artificial Intelligence is to create systems that embody commonsense knowledge in a reliable enough form that it can be used for reasoning in novel situations. Knowledge Infusion is an approach to this problem in which the commonsense knowledge is acquired by learning. In this paper we report on experiments on a corpus of a half million sentences of natural language text that test whether commonsense knowledge can be usefully acquired through this approach. We examine the task of predicting a deleted word from the remainder of a sentence for some 268 target words. As baseline we consider how well this task can be performed using learned rules based on the words within a fixed distance of the target word and their parts of speech. This captures an approach that has been previously demonstrated to be highly successful for a variety of natural language tasks. We then go on to learn from the corpus rules that embody commonsense knowledge, additional to the knowledge used in the baseline case. We show that chaining learned commonsense rules together leads to measurable improvements in prediction performance on our task as compared with the baseline. This is apparently the first experimental demonstration that commonsense knowledge can be learned from natural inputs on a massive scale reliably enough that chaining the learned rules is efficacious for reasoning.

ILP has been successfully applied to a variety of tasks. Nevertheless, ILP systems have huge time and storage requirements, owing to a large search space of possible clauses. Therefore, clever search strategies are needed. One promising family of search strategies is that of stochastic local search methods. These methods have been successfully applied to propositional tasks, such as satisfiability, substantially improving their efficiency. Following the success of such methods, a promising research direction is to employ stochastic local search within ILP, to accelerate the runtime of the learning process. An investigation in that direction was recently performed within ILP [ ˇ Zelezn´y et al., 2004]. Stochastic local search algorithms for propositional satisfiability benefit from the ability to quickly test whether a truth assignment satisfies a formula. As a result, many possible solutions (assignments) can be tested and scored in a short time. In contrast, the analogous test within ILP—testing whether a clause covers an example—takes much longer, so that far fewer possible solutions can be tested in the same time. Therefore, motivated by both the sucess and limitations of the previous work, we also apply stochastic local search to ILP but in a different manner. Instead of directly applying stochastic local search to the space of firstorder Horn clauses, we use a propositionalization approach that transforms the ILP task into an attribute-value learning task. In this alternative search space, we can take advantage of fast testing as in propositional satisfiability. Our primary aim in this paper is to reduce ILP run-time. The standard greedy covering algorithm employed by most ILP systems is another shortcoming of typical ILP search. There is no guarantee that greedy covering will yield the globally optimal hypothesis; consequently, greedy covering often gives rise to problems such as unnecessarily long hypothesis with too

Abstract—Logic based methods for learning from structured data is limited w.r.t. handling large search spaces, preventing large-sized substructures from being considered by the resulting classifiers. A novel approach to learning from structured data is introduced that employs a structure transformation method, called finger printing, for addressing these limitations. The method, which generates features corresponding to arbitrarily complex substructures, is implemented in a system, called DIFFER. The method is demonstrated to perform comparably to an existing state-of-art method on some benchmark data sets without requiring restrictions on the search space. Furthermore, learning from the union of features generated by finger printing and the previous method outperforms learning from each individual set of features on all benchmark data sets, demonstrating the benefit of developing complementary, rather than competing, methods for structure classification. Keywords—Machine learning, Structure classification, Propositionalization.

This paper addresses the role that novel concepts play in learning good theories. To concretize the discussion, I use Hinton’s kinship dataset as motivation throughout the paper. The standpoint taken in this paper is that the most compact theory that describes a set of examples is the preferred theory—an explicit Occam’s Razor. The kinship dataset is a good test-bed for thinking about relational concept learning because it contains interesting patterns that will undoubtedly be part of a compact theory describing the examples. To begin with, I describe a very simple computational level theory for inductive theory learning in first-order logic that precisely states that the most compact theory is preferred. In addition, I illustrate the obvious result that predicate invention is a necessary part of any system striving for compact theories. I present derivations within the Inductive Logic Programming (ILP) framework that show how the intuitive theories of family trees can be learned. These results suggest that encoding regular equivalence directly into the training sets of ILP systems can improve learning performance. To investigate theories resulting from optimization, I devise an algorithm that works with a very strict language bias allowing all consistent rules to be entertained and explicitly optimized over for small datasets. The algorithm, which can be viewed as a special case implementation of ILP, is capable of learning a theory of kinship comparable in compactness to the intuitive theories humans use regularly. However, this alternative approach falls short as it is incapable of inventing the unary predicate sex to learn a more compact theory. Finally, I comment on the philosophical position of extreme nativism in light of the ability of these systems to invent primitive concepts not present in the training data.

Abstract. Increasing the productivity of simulation-based semiconductor design verification is one of the urgent challenges identified in the International Technology Roadmap for Semiconductors. The most difficult aspect is the generation of stimulus for functional coverage closure. This paper introduces a new Coverage-Directed test Generation (CDG) feedback loop which applies Inductive Logic Programming (ILP) to selected tests and coverage data to induce rules that can be used to automatically direct stimulus generation towards outstanding coverage. The case study documented in this paper shows a significant reduction of simulation time when ILP-based CDG is compared to random test generation. This is an exciting and promising new application area for ILP. 1

Abstract: Pattern recognition, machine learning and artificial intelligence approaches play an increasingly important role in rational drug design, screening and identification of candidate molecules and studies on quantitative structure-activity relationships (QSAR). In this review, we present an overview of basic concepts and methodology in the fields of machine learning and artificial intelligence (AI). An emphasis is put on methods that enable an intuitive interpretation of the results and facilitate gaining an insight into the structure of the problem at hand. We also discuss representative applications of AI methods to docking, screening and QSAR studies. The growing trend to integrate computational and experimental efforts in that regard and some future developments are discussed. In addition, we comment on a broader role of machine learning and artificial intelligence approaches in biomedical research.